Introduction
Candidates over-index on the secure and resilient domains and treat cost as common sense. It isn’t. Design Cost-Optimized Architectures is worth roughly 20% of the AWS Certified Solutions Architect - Associate (SAA-C03) exam — a full fifth of your score — and the questions are rarely “which is cheapest?” in the abstract. They hand you a workload with a usage pattern and four technically valid options, and ask for the one that meets the requirement at the lowest cost. Miss the pricing model that matches the usage pattern and you lose points on questions you thought you understood.
The trap is that every wrong answer works. On-Demand EC2 will absolutely run a steady 24/7 database — it’s just the most expensive way to do it. S3 Standard will store your compliance archives — it’s just five times the price of the class built for that. Cost optimization on the exam is pattern-matching: read the usage signal (steady vs. spiky, interruptible vs. critical, hot vs. cold data) and pick the purpose-built, right-sized, pay-for-what-you-use option.
This guide is a practitioner’s tour of the domain the way the exam thinks about it: cost-optimized compute, storage, databases, and network, plus the cost-management tools AWS expects you to reach for. Each section ends with the exam cue that separates the right answer from a plausible wrong one. If you want the big picture first, start with the AWS Solutions Architect Associate Guide 2026 and the exam domains strategy, then come back here to go deep on cost. This is the fourth domain in the set — the resilient and high-performing guides cover the others.
The Cost-Optimization Mindset on AWS
Before the services, internalize the principles the exam rewards. A cost-optimized architecture:
- Matches the pricing model to the usage pattern — steady load buys commitment discounts, spiky load stays elastic, interruptible load rides Spot.
- Right-sizes continuously instead of provisioning for peak-of-peaks and leaving it there.
- Pays for consumption, not capacity — serverless and on-demand pricing win when load is unpredictable or bursty.
- Moves data to the cheapest tier that still meets the access requirement — hot data on fast storage, cold data on archival.
- Eliminates avoidable data transfer — egress and cross-AZ traffic are silent budget killers.
The exam consistently rewards managed + elastic + committed-where-steady over hand-rolled, over-provisioned, or always-on-at-full-price. Keep that instinct handy.
Cost-Optimized Compute: The EC2 Pricing Models
The single densest cost topic is EC2 purchasing options. Know exactly which usage pattern each one is for:
| Model | Discount vs. On-Demand | Commitment | Best for |
|---|---|---|---|
| On-Demand | Baseline | None | Short-term, spiky, unpredictable, or first-time workloads |
| Savings Plans | Up to ~72% | 1 or 3 yr, $/hour spend | Steady compute where you want flexibility across instance families/regions |
| Reserved Instances | Up to ~72% | 1 or 3 yr, specific instance | Steady, predictable workloads on a known instance type |
| Spot Instances | Up to ~90% | None (can be reclaimed) | Fault-tolerant, interruptible, stateless batch/CI/big-data |
| Dedicated Hosts | — (premium) | Optional | Licensing (BYOL) and compliance requiring physical isolation |
The exam cues, memorized cold:
- “Steady-state, runs 24/7, known usage” → Savings Plans or Reserved Instances. Prefer Savings Plans when the scenario stresses flexibility to change instance types or regions; prefer RIs when it needs a capacity reservation in a specific AZ.
- “Fault-tolerant / can tolerate interruption / stateless batch” → Spot Instances (often via an Auto Scaling group or Spot Fleet), the cheapest by far.
- “Unpredictable, short-term, or brand-new workload” → On-Demand until the pattern is known.
- “Must run on dedicated physical hardware for licensing/compliance” → Dedicated Hosts.
Two more compute levers that show up:
- Right-sizing with AWS Compute Optimizer, which analyzes utilization and recommends smaller/cheaper instance types. “The instances are 10% utilized” → downsize.
- Serverless and Graviton. Lambda and Fargate bill per request/second with zero idle cost — the winning answer when load is spiky or event-driven and “no capacity to pay for when idle” is the requirement. Fargate Spot and Graviton (ARM) instances cut compute cost further when the workload allows.
Cost-Optimized Storage: S3 Classes and Lifecycle
S3 storage classes are a guaranteed appearance. Each trades retrieval speed and minimums for a lower per-GB price:
| Storage class | Best for | Retrieval |
|---|---|---|
| S3 Standard | Frequently accessed, hot data | Instant |
| S3 Intelligent-Tiering | Unknown or changing access patterns | Instant; auto-tiers for you |
| S3 Standard-IA | Infrequent access, needs fast retrieval | Instant (per-GB retrieval fee) |
| S3 One Zone-IA | Infrequent, re-creatable, single-AZ OK | Instant (lower durability) |
| Glacier Instant Retrieval | Archive needing millisecond access | Instant |
| Glacier Flexible Retrieval | Archive, minutes-to-hours retrieval OK | Minutes to 12 hours |
| Glacier Deep Archive | Long-term compliance, rarely accessed | 12–48 hours (cheapest) |
The decisive exam cues:
- “Access pattern is unknown or unpredictable” → S3 Intelligent-Tiering (no retrieval fees, moves objects between tiers automatically). This is the safe default when the question won’t tell you the access frequency.
- “Accessed infrequently but must be available immediately” → Standard-IA (or One Zone-IA if the data is reproducible and single-AZ durability is acceptable).
- “Long-term archive, retrieval in hours is fine, lowest possible cost” → Glacier Deep Archive.
- “Automatically move objects to cheaper tiers as they age” → an S3 Lifecycle policy transitioning objects (e.g. Standard → Standard-IA at 30 days → Glacier at 90 → Deep Archive at 180) and expiring them when no longer needed.
Lifecycle policies are the tool the exam wants for aging data cost-effectively; Intelligent-Tiering is for unknown access patterns. Don’t confuse the two. For the full storage picture, the S3 complete guide for SAA-C03 goes deeper on classes, durability, and features.
Cost-Optimized Databases
Database cost mirrors the compute story — commit for steady, go serverless for variable:
- RDS Reserved Instances discount steady, always-on relational databases just like EC2 RIs. “Production database running 24/7 for years” → Reserved.
- Aurora Serverless v2 scales capacity with load and is the answer for variable or intermittent database workloads where paying for a full instance around the clock is wasteful (dev/test, spiky apps, infrequent workloads).
- DynamoDB capacity modes are a classic cost question: On-Demand for unpredictable/spiky traffic (pay per request, no capacity planning) versus Provisioned (with Auto Scaling) for predictable traffic where you can reserve throughput more cheaply.
- Read replicas vs. caching. If a scenario is bottlenecked on repeated reads, an ElastiCache layer can be cheaper and faster than scaling the database tier — a cost and performance answer (see the high-performing architectures guide).
- Right-tier storage. Move infrequently queried data out of expensive hot databases into S3 + Athena for query-in-place at a fraction of the cost when low-latency isn’t required.
Exam cue shorthand: steady DB → Reserved; variable DB → Aurora Serverless or DynamoDB On-Demand; read-heavy → cache before you scale.
Cost-Optimized Network Architecture
Network cost is the domain candidates forget, and AWS knows it. The rules:
- Data transfer in is free; data transfer out to the internet costs money. Egress-heavy workloads should front content with CloudFront, whose edge caching cuts origin egress and is cheaper per GB than serving directly from S3/EC2.
- Cross-AZ and cross-region traffic costs money. Keeping chatty components in the same AZ (where availability allows) or same region reduces transfer charges — a real trade-off against resilience the exam may test.
- NAT Gateway vs. VPC Endpoints. Routing S3/DynamoDB traffic from private subnets through a NAT Gateway incurs both hourly and per-GB processing charges. A Gateway VPC Endpoint for S3/DynamoDB is free and keeps traffic on the AWS backbone — the correct, cheaper answer whenever private resources talk to S3 or DynamoDB. Interface Endpoints (PrivateLink) similarly avoid NAT for many other services.
- Right-size the NAT footprint. Consolidating NAT where AZ-level resilience isn’t required, or replacing it with endpoints, is a common “reduce cost” solution.
Exam cue: “private subnet needs S3 access at lowest cost” → Gateway VPC Endpoint, not a bigger NAT Gateway. “High egress to global users” → CloudFront.
The Cost-Management Toolkit
Finally, know which tool does what — the exam asks you to name the right one for a stated goal:
| Tool | Use it when the goal is… |
|---|---|
| AWS Cost Explorer | Visualize and analyze past/forecasted spend and trends |
| AWS Budgets | Set spend/usage thresholds and alert (or act) when exceeded |
| AWS Cost and Usage Report (CUR) | The most granular line-item billing data for deep analysis |
| AWS Compute Optimizer | Right-sizing recommendations for EC2, EBS, Lambda, ASGs |
| AWS Trusted Advisor | Automated checks flagging idle/underused resources and savings |
| Cost Allocation Tags | Attribute cost to teams/projects/environments |
Exam cues: “alert me before I exceed a budget” → AWS Budgets; “analyze and forecast spending trends” → Cost Explorer; “find underutilized resources to downsize” → Compute Optimizer or Trusted Advisor; “break costs down by team/project” → Cost Allocation Tags.
Putting It Together: A Worked Exam Pattern
Read this the way you’d read a real question. “A company runs a batch image-processing job nightly. The work is stateless, can be retried, and must finish by morning. Cost is the priority. Which compute option?”
Walk the signals: stateless + can be retried (interruption-tolerant) + cost is the priority → Spot Instances, likely in an Auto Scaling group. On-Demand works but costs more; Reserved makes no sense for a job that runs a few hours a night; Savings Plans don’t fit a non-steady pattern. The only signal that matters for the answer is “interruptible + cheapest,” and Spot owns that quadrant.
That’s the whole domain in miniature: identify the usage signal, map it to the purpose-built cost option, and ignore the technically-correct-but-pricier distractors. This maps directly onto the Cost Optimization pillar of the Well-Architected Framework, which the exam draws on throughout.
Practicing Cost Questions Under Exam Conditions
Cost-optimization questions are deceptively easy to read and easy to miss — the distractors are all plausible, and the difference between a pass and a fail is recognizing the usage signal fast, dozens of times, under a clock. You build that reflex by drilling scenario questions until “stateless nightly batch” instantly triggers “Spot” without deliberation.
The AWS Certified Solutions Architect - Associate (SAA-C03) Mock Exam Bundle on Sailor.sh gives you scenario-based questions across all four domains — including the cost trade-offs above — with explanations that reinforce why the cheapest right answer beats the plausible wrong ones. Combined with the SAA-C03 study plan and the common mistakes to avoid, timed practice turns the cost domain from a guessing game into free points.
Conclusion
The Cost-Optimized Architectures domain is 20% of the SAA-C03 exam, and every question reduces to the same move: read the usage pattern, then pick the purpose-built, right-sized, pay-for-what-you-use option — and reject the ones that merely work. Steady compute buys Savings Plans or Reserved Instances; interruptible batch rides Spot; unknown-access data goes to Intelligent-Tiering; aging data flows through lifecycle policies to Glacier; private subnets reach S3 through free Gateway Endpoints, not costly NAT Gateways; and you name the right cost tool — Budgets to alert, Cost Explorer to analyze, Compute Optimizer to right-size.
Internalize the exam cues in each table and the domain becomes some of the most learnable points on the test — the patterns repeat, and once you see them, the “cheapest correct” answer jumps out. From here, complete the domain set with the resilient and high-performing architecture guides, then slot cost into your full timeline with the SAA-C03 study plan.
FAQ
What percentage of the SAA-C03 exam is cost optimization?
The Design Cost-Optimized Architectures domain is roughly 20% of the SAA-C03 exam — the fourth and smallest of the four scored domains, but still a fifth of your total score. It’s enough to decide a pass/fail on its own, so it’s worth deliberate study rather than treating it as common sense.
What’s the difference between Savings Plans and Reserved Instances?
Both offer up to ~72% off On-Demand for a 1- or 3-year commitment. Savings Plans commit you to a dollar-per-hour compute spend and apply flexibly across instance families, sizes, and regions (and to Lambda/Fargate). Reserved Instances commit to a specific instance configuration and can provide a capacity reservation in an AZ. On the exam, choose Savings Plans when flexibility is stressed and RIs when a specific capacity reservation is required.
When should I use Spot Instances on the exam?
Whenever the workload is fault-tolerant, stateless, or interruption-tolerant and cost is the priority — batch processing, CI/CD, big-data, and rendering are classic fits. Spot offers up to ~90% savings but can be reclaimed with two minutes’ notice, so it’s wrong for anything that can’t tolerate interruption, like a stateful production database.
Which S3 storage class should I pick when the access pattern is unknown?
S3 Intelligent-Tiering. It automatically moves objects between access tiers based on usage with no retrieval fees, so you don’t pay a penalty for guessing wrong. Use explicit classes (Standard-IA, Glacier) only when the question tells you the access frequency; use Lifecycle policies when data predictably ages and should transition to cheaper tiers over time.
How do I reduce data transfer costs in a VPC on the exam?
For private resources accessing S3 or DynamoDB, use a Gateway VPC Endpoint — it’s free and avoids NAT Gateway per-GB charges. Front internet-facing content with CloudFront to cut egress, and keep chatty components in the same AZ/region to avoid cross-AZ transfer fees. “Lowest-cost S3 access from a private subnet” is almost always a Gateway Endpoint.
Which AWS tool should I use to get alerted about spending?
AWS Budgets — it lets you set cost or usage thresholds and triggers alerts (or automated actions) when you approach or exceed them. Use Cost Explorer to analyze and forecast trends, the Cost and Usage Report for granular line-item data, and Compute Optimizer or Trusted Advisor to find right-sizing opportunities.